Looking to get into chatbots but don’t want to get burned by a bad ROI project?
First, let us address the elephant in the room. Have you ever asked yourself this question?
Do I need to start learning about chatbots since they will soon replace mobile apps?
As it turns out, the best answer for this is you are asking the wrong question. That is, you don’t need to learn about chatbots because they will replace mobile apps. You need to learn about chatbots because they are a genuinely useful new piece of technology which can improve the user’s experience for some selected use cases.
But first, let us dispense with the idea of how you need to build a chatbot instead of a mobile app. A better idea would be to not build a mobile app in the first place.
Take a look at this excerpt from Patrick McKenzie’s podcast. Patrick is widely regarded as one of the savviest marketers who also understands technology.
If you thought mobile apps on the App Store and Google Play were a rat race, just wait until you see what Google and Facebook (and Slack and Telegram and Kik and what not) have planned for you in the bot stores. If you thought the process of improving your rank on the App Store is difficult, wait until you realize that the process of finding your rank for Actions on Google will be even harder. 🙂
So let us just talk about chatbots, minus the marketing channel.
But first, I encourage you not to build bots which compete with other bots for the user’s attention. If despite my best efforts to dissuade you, you are going to proceed with a bot which you will submit to one of the bot directories, and you have no clear strategy for making money…well at least I tried!
So let us consider use cases which have nothing to do with mobile, except perhaps for the fact that your service will be consumed on a mobile device (i.e. you will build a web based service which incidentally also happens to be mobile friendly).
As it happens, some of these use cases are genuinely hard to achieve using just web based apps.
A while back, on quite a whim, I decided to build a chatbot which will provide a more human friendly interface to the cricket stats available on the Cricinfo website. While doing this, I realized that if my chatbot became powerful enough, I would rather use the chatbot than ANY variation of the web based interface, which happens to look like this:
I had to use the Zoom function on my browser to capture the screenshot, and I still couldn’t zoom out enough before the page formatting got messed up. Thats how many facets/options you have on that page.
Compare it with asking a question such as:
What is Rohit Sharma’s highest ODI score?
And receiving the following response:
Rohit Sharma’s highest ODI score is 264
Now, I didn’t build out the bot to complete functionality, but I know that this is completely doable.
This is a genuine use case where the chatbot outperforms the web app experience, no matter how much more thoughtfully the folks over at Cricinfo can design the UI.
The query page like the one you see above usually uses a search library such as Solr which has this concept of facets. (I don’t know if Cricinfo uses Solr, but many big organizations use it for similar purposes).
To oversimplify somewhat, facets are the options you can use to filter your query results.
The more complex the query you wish to issue, the more options and hence facets you will be adding to your query. At that point, if you can figure out a good way to translate the user’s question from natural language to the correct facet, you can build nice chatbots which can provide an immense boost to productivity (read as: high ROI if used by people who earn high hourly rates, such as managers).
In conclusion, I would ask you to look into using chatbots for building better search query interfaces which utilize natural language.
I am self employed right now, but if I were working at a company I will go and find the reporting web app used most frequently by the top managers.
- Find out what data this reporting app queries
- Capture the intents of the top 5 queries
- Provide a natural language search box which is built on top of a bot framework such as API.AI or Microsoft LUIS where your manager can type queries such as “How many widgets did John sell month to date?” followed by “How does it compare to the previous month?”.And learn how to translate the “it” of the second question properly.
- Extract the ‘params’ from the intent, and send them over to your stored procedures
- Present it as a ‘beta’ option for your managers on the same reporting page on a different tab and also let them know that the accuracy is not 100% (i.e. don’t make decisions based on the results)
- Iteratively improve the accuracy of the reporting until you reach a satisfactory level
Do this as your pet project after hours, and you would have an excellent understanding of some of the capabilities of today’s chatbots at a fairly low risk.